Optimization enabled Deep Quantum Neural Network for weed classification and density estimation

Abstract

Weed classification and detection is an important and critical step in the area of weed control. As weeds exist in all fields and in all seasons, it is required to perform weed control for attaining profitable and optimal crop yields. Due to random spacing in plants, identifying and classifying the weeds in plantation areas results in more complexes. Also, detecting the weeds more accurately is a challenging task. Hence, this research designed a framework using Coot Political Optimization based Deep Quantum Neural Network (CPO-based DQNN) model for weed classification. The devised scheme finds the weed categories accurately at the right time in order to make efficient weed control. The input image is pre-processed and they are segmented using Pyramid Scene Parsing Network (PSPNet). The PSPNet is trained by the CPO algorithm. Based on segmented image results, image augmentation is done. Thereafter, weeds are classified using DQNN that is trained using the CPO algorithm. Finally, the density of weeds is estimated based on the ratio of weed pixels in the image to the total count of pixels present in the image. The outstanding performance attained by the developed approach by the measures of sensitivity, specificity, and accuracy is 0.957, 0.906, and 0.936 for the groundnut weed dataset.